Fast, Nonlinear Phase Estimation with the Non-Modulated Pyramid Wavefront Sensor at Low Strehl Ratio

TL;DR

This paper introduces a nonlinear phase estimation algorithm for pyramid wavefront sensors that operates without modulation, improving accuracy at low Strehl ratios and enabling faster adaptive optics corrections.

Contribution

It presents a novel nonlinear optimization method for PyWFS that increases dynamic range without modulation and is suitable for parallel processing.

Findings

Nonlinear estimator outperforms linear at Strehl ratios above 0.2.

At Strehl 0.4, nonlinear error standard deviation is 0.27 radians.

Algorithm is computationally efficient for modern parallel systems.

Abstract

Most adaptive optics (AO) systems using pyramid wavefront sensors (PyWFS) to estimate the phase of the pupil field use mechanical modulation of the beam in order to increase the dynamic range in low-order modes so the PyWFS can usefully operate at low Strehl ratio. The tradeoff for this approach is reduced sensitivity, which, in turn, makes it difficult to attain a high Strehl ratio once the loop has been closed. We propose an algorithm that increases the dynamic range of the PyWFS without modulation. The proposed algorithm achieves this in two ways: 1) it allows the PyWFS to be treated with any desired optical modeling algorithms, and 2) it employs Newton's method for nonlinear optimization to create an estimator that is more accurate than the corresponding linear estimator. Numerical simulations show that nonlinear optimization can make more accurate estimates of the phase of the wavefront than the corresponding linear estimator for Strehl ratios of the input beam that are greater than about 0.2. As the input Strehl ratio increases, so does the advantage of the nonlinear estimator over the linear one. For example, when the input beam had a Strehl ratio of 0.4 (corresponding to a standard deviation of the phase of about 0.96 radians), the linear estimator error had a standard deviation of about 0.65 radians, while the nonlinear estimator error had a standard deviation of about 0.27 radians (this depends only weakly on the noise level, assuming there is enough signal for the PyWFS to work in the linear regime). The new algorithm can be implemented in massively parallel modes, since the required calculations have almost no inter-dependencies. It is suggested that the required computations can be performed quickly enough for the purposes of AO on modern computer systems.